A Lightweight Deep Learning Model for Human Activity Recognition on Edge Devices

Agarwal, Preeti; Alam, Mansaf

doi:10.1016/j.procs.2020.03.289

Cited by 111 publications

(54 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The activities were recorded with a tri-axial accelerometer sensor. The training, validation, and evaluation splits for the WISDM dataset are adopted from [ 20 , 29 ]. Users 1–24 form training data, 24 and 25 form the validation data, and 26–36 are used for testing.…”

Section: Discussionmentioning

confidence: 99%

“…While the deep-learning-based methods rely on a fixed window size to extract temporal sequences from time-series sensor data, DTE uses a number of different window sizes as input and trains a neural network ensemble. This helps boosting the classification metrics when compared to some previous works [ 6 , 20 , 29 , 30 ]. Furthermore, DTE can be used with any base neural network architecture.…”

Section: Related Workmentioning

confidence: 98%

See 1 more Smart Citation

Confidence-Calibrated Human Activity Recognition

Roy

Girdzijauskas

Socolovschi

2021

Sensors

View full text Add to dashboard Cite

Wearable sensors are widely used in activity recognition (AR) tasks with broad applicability in health and well-being, sports, geriatric care, etc. Deep learning (DL) has been at the forefront of progress in activity classification with wearable sensors. However, most state-of-the-art DL models used for AR are trained to discriminate different activity classes at high accuracy, not considering the confidence calibration of predictive output of those models. This results in probabilistic estimates that might not capture the true likelihood and is thus unreliable. In practice, it tends to produce overconfident estimates. In this paper, the problem is addressed by proposing deep time ensembles, a novel ensembling method capable of producing calibrated confidence estimates from neural network architectures. In particular, the method trains an ensemble of network models with temporal sequences extracted by varying the window size over the input time series and averaging the predictive output. The method is evaluated on four different benchmark HAR datasets and three different neural network architectures. Across all the datasets and architectures, our method shows an improvement in calibration by reducing the expected calibration error (ECE)by at least 40%, thereby providing superior likelihood estimates. In addition to providing reliable predictions our method also outperforms the state-of-the-art classification results in the WISDM, UCI HAR, and PAMAP2 datasets and performs as good as the state-of-the-art in the Skoda dataset.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 98%

Confidence-Calibrated Human Activity Recognition

Roy

Girdzijauskas

Socolovschi

2021

Sensors

View full text Add to dashboard Cite

show abstract

“…Many researchers have made great efforts in this regard. Agarwal et al [28] proposed a lightweight deep learning model for HAR and deployed it on Raspberry Pi3. This model was developed using a shallow RNN in combination with the LSTM algorithm, and its overall accuracy on the WISDM dataset achieved 95.78%.…”

Section: Related Workmentioning

confidence: 99%

LSTM-CNN Architecture for Human Activity Recognition

2020

View full text Add to dashboard Cite

In the past years, traditional pattern recognition methods have made great progress. However, these methods rely heavily on manual feature extraction, which may hinder the generalization model performance. With the increasing popularity and success of deep learning methods, using these techniques to recognize human actions in mobile and wearable computing scenarios has attracted widespread attention. In this paper, a deep neural network that combines convolutional layers with long short-term memory (LSTM) was proposed. This model could extract activity features automatically and classify them with a few model parameters. LSTM is a variant of the recurrent neural network (RNN), which is more suitable for processing temporal sequences. In the proposed architecture, the raw data collected by mobile sensors was fed into a two-layer LSTM followed by convolutional layers. In addition, a global average pooling layer (GAP) was applied to replace the fully connected layer after convolution for reducing model parameters. Moreover, a batch normalization layer (BN) was added after the GAP layer to speed up the convergence, and obvious results were achieved. The model performance was evaluated on three public datasets (UCI, WISDM, and OPPORTUNITY). Finally, the overall accuracy of the model in the UCI-HAR dataset is 95.78%, in the WISDM dataset is 95.85%, and in the OPPORTUNITY dataset is 92.63%. The results show that the proposed model has higher robustness and better activity detection capability than some of the reported results. It can not only adaptively extract activity features, but also has fewer parameters and higher accuracy. INDEX TERMS Human activity recognition, convolution, long short-term memory, mobile sensors.

show abstract

“…A lightweight deep learning classifier CNN is shown in Figure 4 , in which

and

are the input layers,

and

are the hidden layers, and

is the output layer.The basic architectures of feed forward network and recurrent neural networks are shown in Figure 3 and Figure 4 . The feed forward network gives accuracy of about 97.4%, and RNN individually is 95.5% accurate [ 16 ].…”

Section: Lightweight Deep Learning Classifier Hybrid Modelmentioning

confidence: 99%

An Efficient and Lightweight Deep Learning Model for Human Activity Recognition Using Smartphones

Ankita

Rani

Babbar

et al. 2021

Sensors

View full text Add to dashboard Cite

Traditional pattern recognition approaches have gained a lot of popularity. However, these are largely dependent upon manual feature extraction, which makes the generalized model obscure. The sequences of accelerometer data recorded can be classified by specialized smartphones into well known movements that can be done with human activity recognition. With the high success and wide adaptation of deep learning approaches for the recognition of human activities, these techniques are widely used in wearable devices and smartphones to recognize the human activities. In this paper, convolutional layers are combined with long short-term memory (LSTM), along with the deep learning neural network for human activities recognition (HAR). The proposed model extracts the features in an automated way and categorizes them with some model attributes. In general, LSTM is alternative form of recurrent neural network (RNN) which is famous for temporal sequences’ processing. In the proposed architecture, a dataset of UCI-HAR for Samsung Galaxy S2 is used for various human activities. The CNN classifier, which should be taken single, and LSTM models should be taken in series and take the feed data. For each input, the CNN model is applied, and each input image’s output is transferred to the LSTM classifier as a time step. The number of filter maps for mapping of the various portions of image is the most important hyperparameter used. Transformation on the basis of observations takes place by using Gaussian standardization. CNN-LSTM, a proposed model, is an efficient and lightweight model that has shown high robustness and better activity detection capability than traditional algorithms by providing the accuracy of 97.89%.

show abstract

A Lightweight Deep Learning Model for Human Activity Recognition on Edge Devices

Cited by 111 publications

References 26 publications

Confidence-Calibrated Human Activity Recognition

Confidence-Calibrated Human Activity Recognition

LSTM-CNN Architecture for Human Activity Recognition

An Efficient and Lightweight Deep Learning Model for Human Activity Recognition Using Smartphones

Contact Info

Product

Resources

About